“Determinants of international reserve currency issuers in the
light of an altering international monetary system”
Abstract:
Since new economic powers such as China and the Eurozone ascend to the United States, the monopolistic future of the US-Dollar as primary foreign exchange reserve currency is increasingly challenged. This work assesses the determinants of nations, whose currencies are used by other central banks as foreign exchange reserves. Figures of the composition of worldwide foreign exchange reserves (COFER) are contextualized with multiple economic indicators. A dynamic panel model consisting of 11 countries over a time span of nearly 40 years, yielded a positive influence for increasing GDP levels, trade volume, financial depth and military expenditure on the relevance of a national currency as foreign exchange reserves. Likewise, a positive effect of democratic structures and financial openness was found.
Key Words: Foreign Exchange Reserves, Dollar Hegemony, International monetary system
MSc International Economics and Business
Presented by Carlo Schuster
S3781038
c.l.schuster@student.rug.nl
Supervisor: Assistant Prof. Dr. Andreas Steiner
Co-Assessor: Prof. Dr. Dirk Bezemer
2 Index: List of abbreviations 3 1. Introduction 4 2. Literature Review 5 3. Data 11
3.1.Multicollinearity, Heteroskedasticity, Autocorrelation and Stationarity 16
3.2.China as an outlier 18
4. Methodology 19
4.1. Relative or absolute figures 20
4.2.Model equation and hypothesis 21
4.3.Model specifications 22
5. Results and Interpretations 23
5.1.Individual and pooled OLS regressions 23
5.2.Fixed-effects estimation (excluding inertial effect) 27 5.3.Dynamic model estimations (including inertial effect) 30
6. Robustness 33
6.1.Robustness check with time-fixed effects 33
6.2.Robustness check with relative indicators 35
6.3.Robustness check by removing small countries 36
6.4.Additional remarks on robustness 37
7. Conclusion 39
8. Appendix 41
3 List of Abbreviations:
AUD Australian Dollar
BIS Bank of International Settlements
CAD Canadian Dollar
CHF Swiss Franc
CNY Renminbi (Chinese Yuan)
COFER Currency composition of foreign exchange reserves
c.f. compare also
ECB European Central Bank
EUR Euro
GBP Sterling (British Pound)
GDP Gross domestic product
GMM Generalized method of moments
IFS International Financial Statistics
IMF International Monetary Fund
JPY Japanese Yen
M2 Aggregated money supply
OLS Ordinary least square estimator
SDR Special drawing right
UN United Nations
USD US-American Dollar
4 1. Introduction
The power and supremacy of the US and its national currency the US-Dollars remained unchallenged for a long time. Since the world power of the United Kingdom declined and with the installation of the Bretton Woods regime in 1944, the US-Dollar was positioned as the new primary international reserve currency with a fixed exchange rate to the US gold stocks in the Fort Knox treasuries. Worldwide demand for international liquidity skyrocketed, which could not be provided by the US due to insufficient gold reserves; US-American fiscal deficits increased considerably due to financing of the Vietnam war amongst other reasons. Other major players, in particular France, questioned whether the US would be able to maintain parity to gold. Parity was eventually suspended in the early 1970s. National currencies started to float against others and national central banks were able to diversify their international foreign exchange reserves in favor of other currencies such as the Deutschmark or the Japanese Yen. With the end of the cold war the status of the US, as the only economic superpower, was further strengthened. With the upcoming monetary union in Europe, voices were raised, propagating the Euro as the challenger of the US-Dollar hegemony in the international monetary system (Mundell, 2000). Yet so far, the Euro has failed to meet these expectations. Undoubtedly, China as the new ascending economic power will eventually claim a superior position in the global monetary system and could challenge the US.
Graph 1: Economic figures of 2000 and 2018 of the USA, Eurozone and China in comparison
Graph 1 underlines this evolvement within the last two decades by comparing the COFER1 share, absolute GDP, trade volume, money supply (M2) and military expenditures (as a proxy for geopolitical power) of China, the US and the Eurozone. All variables have been aggregated;
1 Currency Composition of Official Foreign Exchange Reserves 0% 20% 40% 60% 80% 100%
COFER GDP Trade M2 Defense
spending
2000
China Eurozone USA
0% 20% 40% 60% 80% 100%
COFER GDP Trade M2 Defense
spending
2018
5
shares are displayed in Graph 1. In 2000 the US was undoubtedly the most infuential economic superpower, while the at the time created Eurozone similarly as in 2018 accounted for one fourth to one fifth of the aggregated indicators. Whereas back in 2000, China as low-industrialised country fell behind the Eurozone and the US in all indicators, by 2018 it outran the Eurozone in the latter four indicators and approaches the US. In absolute terms, while the latter four indicators for the US and the Eurozone doubled to tripled within twenty years, the Chinese indicators increased more than tenfold (Prasad, 2017).
The US-American fiscal debt levels rise continuously to an unforeseen extent. China among other Asian economies have thus become major creditors to the US. As the credibility in the debt issuing country and the repayment prospect of the issued debt deteriorates considerably with rising debt levels, the question raises to what extent US-American sovereign debt will remain unchallenged. Therefore, the question for an appropriate successor for the US-Dollar in the foreseeable future seems evident.
Already in 1969, the International Monetary Fund (IMF) introduced with its Special drawing rights (SDR), an artificially currency, which is aggregated out of several national currencies within a currency basket. Prior to the suspension of Bretton Woods, concerns were raised (Brittan & Triffin, 1967), that the US would no more be able to maintain the Gold to US-Dollar parity due to shortages in gold. While, SDR could have occupied the successor’s role, its relevance as a means of settlement or investment was always immaterial. Alternatively, central banks usually rely on gold and foreign exchange, which usually takes the shape of long-term government bonds of the respective country. Seeing as aggregated currencies based on currency baskets, such as SDRs, have failed to succeed the US-Dollar, their potential as a successor will not be further discussed in this thesis. Instead, this paper aims at assessing the characterizing determinants of countries, providing the strength, superiority and ability of its national currency to internationally circulate and challenge the US-Dollar as the major international reserve currency.
2. Literature Review
US-6
Dollar as international reserve currency lies in its extensive capital markets, trading assets denominated in the greenback (Kindleberger, 1967).
Barry Eichengreen and Jeffrey Frankel, whose research is focused on international reserves and questions concerning the international monetary system, wrote a chapter for “The Future of the SDR in Light of Changes in the International Monetary system”, a publication of the International Monetary Fund in 1996. At the time the international monetary system was still mostly unipolar, however the Maastricht Treaty in 1992 paved the way for the inauguration of a new major currency, the euro. Under the impression of a new possible counterpart of the US-Dollar they elaborate different determinants, favoring a currency to become an international reserve currency. They argued that a large share in international output, trade and finance was a key factor in determining the comparative strength of an economy to issue a reserve currency. In favor of its own national currency, economic major players, responsible for a significant share of world output, trade and finance are likely to specify the currency used to invoice international transactions in international commodity markets. Likewise, they stated that money markets needed to fulfill several requirements. First, they must be open and free from capital controls. Second, they must provide enormous amounts of liquidity and therefore must be deep and well developed. Additionally, according to Eichengreen and Frankel the stable value of the reserve currency must be ensured. An erratically fluctuating value daunts international financial agents and other nations to use the currency for their operations. Finally, the two authors describe an inertial bias to remain with the reserve currency previously used (Eichengreen & Frankel, 1996). An equilibrium in the global monetary system is formed in favor of a monopolistic reserve currency. Although key characteristics change over time and the country’s predominant position vanishes, the superior role of the currency in the international monetary system may persist. Network externalities constantly self-reinforce the position, albeit a more optimal alternative might be available. (Krugman, 1980). The establishment of a new international reserve currency has even been described as irreversible and hysteresis (Guidotti & Rodriguez, 1992).
7
(Norrlof, 2014). She also pointed to the necessity of statutory frameworks with strong property rights, deposit protection and a prolonged legal practice capable of providing a safe environment for investment. Moreover, the ability to reinforce and protect the status quo militarily is needful. Additionally, she accented that defense spending and provided protection for allies induces an economic return, since allied countries support the installation of the national currency as international reserve currency with their investment and trust (Norrlof, 2010). Eichengreen and Frankel stressed on the importance of strong central bank and a large financial sector to counterbalance the political influence and demonstrate monetary independence (Eichengreen & Frankel, 1996). Likewise, Ottmar Issing, former chief economist of the European Central Bank (ECB), named the indispensability of reliability and transparency of national money policies making institutions as well as, upholding the rule of law and it operating principles (Issing, 2008). Globally active international agents are briefed and accustomed to the prolonged legal practices in the US and their monetary policy decision making rules.
8
Risk averse market actors refrain from switching and, thereby, imitatively affirm the established practices (Cohen, 2011). Above all it remains evident, that factors such as network externalities, economies of scale and scope, hysteresis and inertia effects are highly influential. At the same time quantifying them or controlling for themremains arduous.
Marc Flandreau and Clemens Jobst added in 2006 a historical analysis to the ongoing work on determinants, empowering the leadership position of a national currency in the international monetary system. With a dataset going back to the 19th century, the two authors analyzed the prime time of the Sterling as global leader in the international monetary system. They were able to find viable proof for the existence of persistence and strategic externalities in favor of the Sterling. These externalities arise in the form of economies of scale and scope. Thereby, the position of a reserve currency issuer is reinforced by lowering transaction costs, increasing synergies and upscaling effects. These effects reinforce themselves and are prone to have a hysteresis effect. Sterling was still in place in the international monetary system, albeit the supremacy of the United Kingdom had slowly vanished. Moreover, they attached greater importance to an economy’s trade share, as a considerable driver of international monetary leadership (Flandreau & Jobst, 2006).
9
(majorly Japan, Korea, Saudi-Arabia and Germany) and opposing independent military powers (for instance China, Russia or India). For the first group their international reserve investment is majorly or exclusively denominated in US-Dollar, as previously also described by Carla Norrlof (Norrlof, 2010). Whereas, the latter are nuclear powers themselves and, hence, not as prone to US-American foreign and security policies. Their scenarios elaborated the conse-quences of a foreign and militarily disengagement of the US, to its national economy (Eichengreen, Mehl, & Chitu, 2017).
In the “The Exorbitant Burden” (2016), Ganziro and Vambery provided an assessment of the relationship between the leadership position of the US-Dollar in the international monetary system and the supremacy of the US in the world order to the deficits in its balance of payments and trade. In their econometric analysis they considered numerous indicators which display determinants substantiating the leadership position of a reserve currency issuer in the international monetary system. The analysis and conclusions revealed a clearly pro-American impression, suggesting a picture of an unwavering US-American supremacy in the international monetary and political order. As no other country is able to provide the conditions qualifying the US as single international reserve currency issuer, today and in the foreseeable future. Apart from the US, no other country is featured with such a deep and extensive financial market providing these enormous amounts of liquidity, no other country is geopolitically as powerful with a military strength able to ensure worldwide trade and if necessary to defend the supremacy in the international monetary system. The main purpose of their work was the analysis of the consequences for the US-American interest rates on their long-term sovereign bonds, when the US-American supremacy ceases. However, they also admitted that the development of the US sovereign interest rates often follows a countercyclical evolvement. Most notably, in the after-math of the financial crisis, which was caused by excessive subprime mortgage lending in the United States, US-American sovereign interest rates sank, although the fiscal deficit skyrocketed. While one might assume increasing risk premia, when global uncertainty increases and the fiscal deficits increases considerably, in a recession dissuaded investors refrain from increased risk taking and return to the safe haven, the known treasuries denominated in US-Dollars (Ganziro & Vambery, 2016). At the same time, the eruption of the European crisis impeded the potential of the Euro as a challenger of the dollar hegemony durably.
10
international currencies and differentiated for private and official use (Cohen, 2015).2 Although this work focuses on the role of international currencies as foreign exchange reserves (displayed as bold in the schema 1), the other functions should not be left out. Thereby not only sovereign actors trigger national currencies to evolve to international reserve currencies. Although these are factors, which cannot directly be attributed to nations or their government. Instead financial agents from minor economies support the prevalence of international reserve currencies as its acceptance is guaranteed universally, by using them as a vehicle in settlements within two different small currencies. Likewise, enormous amounts of foreign exchange by private actors flow back to its national financial market in form of investment. (Cohen, 2015; Norrlof, 2014). Thereby significant economies of scale and scope come into effect.
Schema 1 discloses the motives of central banks to hold and accumulate foreign exchange reserves. Firstly, they are used to as an intervention currency, to stabilize the value of the own national currency in intervention to the foreign exchange market. Secondly, the value of international currencies is employed as anchor currency to stabilize the national currency. The third motive encompasses the store of value function of money. Central banks deposit foreign exchange reserves to back up the value of their own issued currency (see per example Prasad, 2017).
The chief economist of the IMF Gita Gopinath jointly elaborated with Jeremy Stein possible scenarios in which the US-Dollar loses ground with respect to other currencies as leading international foreign exchange. Although the economic size and the trade volume of the US and Eurozone are on a par in absolute numbers, the volume of international trade invoiced in Euro with a factor of 1.2 rarely exceeds the Eurozone’s imports. The amount of international trade invoiced in US-Dollar, in contrast, exceeds US-American imports by a factor of 4.7 (Gopinath, 2015). If global trade flows shift to different continents and the USA reinforces its protectionist trade policies, the global supremacy of the US-Dollar will lessen (Gopinath & Stein, 2018). The share of the Renminbi has gained enormous relevance, increasing from less than 1 percent in 2010 up to 25 percent in 2015 in Chines trade settlement. Recently, the Renminbi has taken over the Euro as second most used currency in global trade finance (Eichengreen, Mehl, & Chiţu, 2017).
11 3. Data
Different international organizations, most notably the International Monetary Fund (IMF), the World Bank as well as the Organization for Economic Co-operation and Development (OECD) maintain extensive data bases providing specific country data. Data coverage on international reserves held by individual central banks and all of them globally is only partially available. As the composition and amounts of international reserves held by central banks are crucial for the stability of national financial systems and the maintenance of currency boards and pegs, several ones are reluctant to publish any data, among them major reserve holders such as Taiwan, Saudi-Arabia, Indonesia or Qatar. From 1995 until 1998 in a yearly interval and thereafter quarterly the IMF has gathered data on the composition of the world wide international foreign exchange reserves by central banks worldwide in its Currency Composition of Official Foreign Exchange Reserves (COFER) database. COFER includes the US-Dollar, Euro3 (and its most prominent precursors Deutschmark, French Franc, Dutch Guilder), Japanese Yen, British Pound, Chinese Renminbi (since 2016), Canadian and Australian Dollars (since 2012) and Swiss Franc. Within the years of observation these currencies were responsible for more than 95 percent of all worldwide foreign exchange reserves held by central banks. Recently, a distinction between central banks of advanced as well as of emerging and developing countries has been introduced. However, discrepancies among the two indicators are minor and data coverage is limited. Therefore, an integration of the differentiation would be devoid of significance.
However, not all central banks report data to the IMF for their COFER statistics. As the total amount of reserves can be consulted in the International Financial Statistics (IFS), likewise provided from the IMF, the coverage (percentage of allocated reserves) of COFER can be assessed. The biggest foreign exchange reserve holder China started contributing data for COFER only five years ago. Only a few central banks, such as Russia, Brazil or Switzerland, are completely transparent and publish the allocation to individual currencies of their foreign exchange reserves (Banco Central do Brasil, 2018). Yet, numbers are only published unregularly, and long-lasting reliable time series are not available.
12
Graph 2: Coverage of COFER of all global foreign exchange reserves, IMF
Graph 2 depicts the coverage of COFER among the global foreign exchange reserves. Thereby, around the turn of the millennium almost 80 percent of all foreign exchange reserves could be allocated to a specific currency. Subsequently the coverage reduced considerably, down to 54 percent in the end of 2014. Since then China and other reserve holders started reporting their composition to the IMF. Lately, more than 93 percent of foreign exchange reserves can be allocated to a specific currency. The data coverage of COFER is not optimal, as specific countries refuse to publish data. Nonetheless, COFER is still the most extensive and precise database providing information concerning the composition of central banks worldwide. Although the absolute volume of international reserves held by central banks is known from the IFS published by the IMF, these shortcomings in data coverage regrettably give rise to a selection bias problem. If for instance the numerous countries, which refused to report their data on their reserve composition to the IMF, share the specific characteristics of opposing the Dollar as reserve currency due to foreign policy conflict with the US, the share of the US-Dollar can be highly overestimated. Still, from the overview of the various data sources for international foreign exchange reserves and their allocation to individual currencies provided by Philip Woolridge from the Bank of International Settlements (BIS) (Wooldridge Philip D., 2006), the COFER database remains the optimal choice.
To enhance the validity of the performed econometrical analysis the observation period was complemented with currency composition of the global foreign exchange data, which was
10 20 30 40 50 60 70 80 90 100 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 Perc en ta ge o f allo cat ion to a speci fic cu rre n y
13
published by the IMF in their annual reports. From the reports of 1990 (IMF, 1990) and 1995 (IMF, 1995) historical data was retrieved extending data coverage back to the year of 1980. Graph 34 displays the share of different currencies among worldwide foreign exchange reserves held by central banks worldwide. The observation period includes the year of 1980 until the end of 2018. The most prominent investment currency has always been the US-Dollar fluctuating within a ratio of 57 until 72 percent of the worldwide held foreign exchange reserves. Apart from the greenback, until the introduction of the Euro in 1999, its key precursors, most notably the Deutschmark was ranging in low double figures, whereas the French Franc and the Dutch Guilder were only of minor importance reaching values below five percent until 1998. The latter three named currencies merged into the Euro in 1999, pooling their strength. In the following years, the Euro increased its share from 18 percent in 1999 to approximately 28 percent by 2010. With the eruption of the European debt crisis and the slower recovery of the global financial crisis of 2008 with respect to the United States, the relevance of the Euro as an international foreign exchange reserve diminished to the range of its initial value of 1999. Interestingly, after the inception of the global financial crisis in the United States the relevance of the common currency increased with respect to the greenback. It was not until 2011 that its relevance sank. Besides the two currencies on the shore of the Atlantic, the Japanese Yen has played a vital role among the international reserve currencies since the mid-1980s. However, since its primetime in the 1980s and early 1990s, where the ratio reached high single digit values, the Japanese Yen’s relevance has slowly but continuously diminished. This development falls into the period of their two lost decades, which are characterized by stagflation and low growth rates. The relevance of the British Sterling as precursor of the US-Dollar as most prominent international reserve currency, is still given 75 years after its replacement during the conference of Bretton Woods. Its values range between 3 and 5 percent. The last currency which has always been reported as international reserve currency by the IMF is the Swiss Franc. However, since the mid-1990s its relevance is vanishingly low (ratios of 0.2 percent) and unchanged. Therefore, omitting the Swiss Franc might be suggestive as no changes can be observed and explained with the regression analysis. This can potentially increase the validity of the regression analysis. Lately, new currencies have gained importance among the international reserve currencies. In 2012 the Australian and Canadian dollar have been added to the COFER database. Since then, their ratios were almost identical, ranging at circa two
14
percent each. More recently in 2016, the Chinese Renminbi was added to the COFER database. Its values in 2018 also ranged at round about two percent.
Above all the numbers in Graph 3 highlight that the Euro as a possible competitor of the US-Dollar failed to gain sustainable influence. Although its performance was promising within the first ten years, it subsequently diminished in influence on its initial values since 2014. Also, the figures fail to display any sudden or volatile changes in COFER. As this is not given, less changes can be explained by the following regression analysis, which reduces the validity of the analysis. By expanding the observation period back to 1980 however, the amount of changes and, therefore, the validity of the analysis can be increased.
COFER indicates the relevance of a reserve currency by stating the sum of the worldwide claims (measured in US-Dollars) in the specific currency or as a share of overall worldwide foreign exchange reserves. It should be noted that this number only includes reserves that can be allocated to a specific currency.5 As the absolute amount of reserves quintupled within 10 years,
it is advisable to consult the percentage shares of the composition of foreign exchange reserves. Otherwise it is necessary to control for the skyrocketing accumulation of foreign exchange reserves of majorly Asian central banks in the 2000s (Ben Bernanke, 2005). Therefore, the COFER share will be used as dependent variable in the executed regression analysis.
Apart from China all considered countries are OECD members. Beneficially, the database of OECD provides quarterly numbers for multiple economic factors for all its members and other countries including China. As economic determinants characterizing a reserve issuer GDP (output approach, noted in millions of USD from IMF) are considered in two different variants. Firstly, the absolute amount of GDP, additionally, a proportionate measure (GDP per capita) will be used as a robustness check. Both have been derived from the OECD database. Moreover, the world bank maintains an extensive data base with numerous indicators for this analysis. Numbers on defense spending by country have been extracted from the latter. Yet, only annual numbers are available from this database. Therefore, for this analysis the annual numbers have been distributed equally upon the quarters of the corresponding year, to increase data coverage. Quarterly trade account data for specific countries were derived from the International Financial Statistics (IFS) database of the IMF. By implementing the money supply (M2), the financial depth and breadth of a country is controlled for (Eichengreen, Mehl, & Chitu, 2017.1). However, admittedly throughout the world there are minor divergences on the definition of
15
money aggregates,6 which have been disregarded. Numbers, originally given in national currencies; converted to US-Dollars were extracted from the database of the Federal Reserve Bank of St. Louis.
The wide use of a currency in international financial markets is key to its emergence as a leading international foreign exchange reserve currency. Since the international acceptance of third party country’s currencies is limited, they rely on vehicle currencies as intermediary to settle international payments (Cohen, 2015). The Bank of International Settlements (BIS) publishes in its triennial survey figures on the use of currencies as vehicles on global markets (BIS, 2016). These figures are considered as explanatory variable in the dataset.
Within the observation period incisive crises hit countries in the sample. The global financial crisis of 2007 and 2008 had a major impact. Whereas output levels sank, money aggregates increased subsequently with intervention policies, such as quantitative easing programs. The latter crisis induced the outbreak of the European debt crisis which undermined the economic status of the Eurozone, with diminishing output, depreciating currency, low inflation levels and increasing money supply. To account for the effect of these two crises dummy variables have been added to the sample to correct for the dynamics of these crises. As previously outlined, the accountability and the operating principles of a central bank are indispensable to raise attractiveness for a reserve currency among central bankers. Monetary policies and in-dependence of central banks differ throughout the world. Price stability however, is a key goal all the central banks in the dataset try to achieve. The inflation as a measure for price stability and strength of the economy is included in the analysis as a proxy for the credibility and accountability of the central bank’s monetary policy (Eichengreen et al., 2016). Moreover, the variable democracy, uses the freedom house index (Freedom House, 2019) of the respective year to introduce the importance of democratic structures to reserve issuing countries. The openness of the capital account is evaluated with the chinn-ito index (Chinn & Ito, 2005). For ordinary investors the expected return from investment is of major concern, when choosing within several finance projects. The central bank as investor for the country also chooses among different currencies, to invest their international reserves in. Therefore, the dataset takes interest rates (3-months or 90-day rates and yields of interbank rates for the specific countries) into consideration. Two major concerns arise with the incorporation of interest rates. Firstly, the interest rate reveals an endogeneity problem. It remains questionable whether the currency is
6 For instance: https://www.federalreserve.gov/faqs/money_12845.htm or
16
globally used due to its low interest rates or whether the interest rates are as low, because the currency is an international reserve currency. To overcome this endogeneity problem an instrumental variable is necessary. Secondly however, analyzing the effect of the interest rate to the restructuring of international reserves is difficult, as different factors come into play. The investment strategies of different countries might diverge with respect to their individual risk-taking and relations to specific foreign exchange reserves issuing countries. Also, all considered countries apart from China can be classified as high-income countries, which are highly intertwined in their monetary policies and are all exposed to a global financial cycle, to which they react by changing their interest rates. Many countries follow the interest rate changes of the FED, as most prominent foreign exchange reserve issuer, causing a bias on other interest rates. Finally, it is not given that higher interest rates automatically translate into a restructuring with respect to the high interest rate economy, as central banks unlike private investors do not invest for yield. The interest rate as an explanatory variable will be only added to the regression with wariness due to these concerns.
Other factors as returning to the safe haven, increasing debt levels, the unlikeliness of these governments to become insolvent and relations within states based on common languages, historical heritage, legal system and colonial ties can be highly relevant (Eichengreen, Mehl, & Chitu, 2017). As the data acquisition on reserve composition of individual central banks is strongly limited to smaller trivial reserve holders, correcting for these individual effects is not feasible. Therefore, these effects must be omitted.7
Japan being trapped in the two lost decades, characterized by a stagflation, has had vanishingly low interest rates around zero percent for the last twenty years, which is an additional factor reducing the validity of this variable with respect to the explanation of restructuring international reserve currencies globally.
3.1.Multicollinearity, Heteroskedasticity, Autocorrelation and Stationarity
The used variables are prone to high correlations among themselves. These can be problematic as they indicate an uprising multicollinearity subject. Collinearity biases estimators and inflates the coefficient of determinations R². The detection of collinearity is assessed by taking the correlation of the explanatory variables. In the present case the variables of absolute GDP, trade volume, financial depth and defense spending exhibit correlations beyond the critical threshold
17
value of 0.88 . Therefore, assessing these variables jointly can be critical as results will be prone to deflated t-statistics and erroneously high R², presuming a high and significant explanatory power. Moreover, in the correlation matrix (A.2.1) the variable vehiclerate displays high correlations with other variables. However unlike the four previously mentioned variables the corresponding variance inflation factor (VIF), yields tolerable values.9 While the first four variables yield excessive VIF values, which ones more underlines the necessity to only assess them individually, the VIF of vehiclerate is tolerable (Wooldridge, 2016). Additionally, the indicators for freedom of the capital account, assessed with the Chinn-Ito-Index (Chinn & Ito, 2005), and democracy, assessed with the Freedom-House-Index exhibit a very high correlation. Apart from China still imposing capital controls in an authoritarian political system, all Western countries perform well in these two indicators. This results in high correlation within these two indicators. Therefore, considering them jointly is problematic, too.
Solely the lagged dependent variable, meant to correct for inertia in the model, is obviously prone to massive multicollinearity. This is a concern, as correlation with numerous other explanatory variables prevails. Since, the absorption of inertia in the assessment is essential for the determination of the characteristic of an international foreign exchange issuer, the high correlations can be connived in this case. However, they require a specific estimator.
An additional arising problem consists of the heteroskedastic dispersion of the dataset. The computed scatterplots10 of the data revealed heteroskedasticity for several variables. This visualized finding was reconfirmed by an executed White-Test.11 A heteroskedastic dispersion of the data violates the assumptions of Gauss-Markow. Robust standards errors can be used to cope with heteroskedasticity at the expense of the significance of the findings.
Since the dataset includes a lagged dependent variable as independent variable to correct for the inertia in the restructuring effect of foreign exchange reserves, the fixed and random effects estimators are biased and therefore not applicable. As we have a dynamic panel data set, we must rely on a GMM-estimator. However, the additional problem of serial correlation of any order might arise, which restricts the choice of GMM estimators. An executed Woolridge-Test12
yielded the absence of first order serial correlation. Consequently, we are not restricted on the
8 for correlation matrix consider the appendix (A.2.1). 9 See Appendix (A.2.3) for the VIF calculations 10 Scatterplots are depicted in the appendix (A.2.4).
18
choice of GMM estimators and can rely on the Arellano–Bover/Blundell–Bond linear dynamic panel-data estimation for our estimations of the dynamic panel model.
Likewise, we must revise whether stationarity is present in our data set. The corresponding executed Fisher type unit root test yielded a rejection of the null hypothesis.13 Since the null hypothesis has been rejected, at least one of the panels is stationary. Additionally, the individually executed regressions of table 1.114 revealed a non-stationary time series if individually regressing few variables. However, in a joint regression, which are solely used to deduce results, the coefficients of the lagged dependent variable outnumber one, which indicates a stationary time series.
3.2.China as an outlier
Although, according to IFS, China is by far the biggest foreign exchange holder, accounting for more than one fourth of worldwide foreign exchange reserves, it has hitherto only played a rather minor role as foreign exchange reserve issuer. Despite China’s increasing weight in the world economy and in the international financial system, COFER only incorporated Chinese Renminbi in 2016. China’s economic figures, which are taken into consideration in the economic analysis, are impressive. For instance, numbers on nominal GDP, trade volume, and money supply are relatively high. China’s COFER ratio has only recently been added and is so far of minor weight. However, China is the only undemocratic country in the sample. Additionally, China still imposes capital controls and the accountability of the People’s Bank of China is comparatively low, as the independence of the central bank is not guaranteed and sudden interventions in the monetary and exchange rate policy are common. As, according to Eichengreen and Frankel, (1996) the freedom of the financial market is inevitable for the functioning as an international reserve currency, the free movement of capital should not be restricted. Chinn and Ito (2005) assessed the strength of financial openness, i.e the absence of capital controls by creating their index. Given that only China imposes capital controls, the introduced Chinn-Ito-Index only displays diverging values for China. Since most countries in the different time periods dispose the maximum value, the measure could be considered a China-specific variable.
The same applies for the democracy index. Unlike all other countries, China is the only non-democratic country in the sample. Apart from some countries in the 1980s, when not all civil
19
liberties were guaranteed, all other countries perform outstandingly in the Freedom House democracy index. China’s scores are constantly poor, with ranking at the bottom of the scale. China, therefore, has to be regarded as an outlier for these two variables.
Recently, China has been easing restrictions on its capital markets, which is also seen in its Chinn-Ito figures. This internationalization of the Renminbi is considered a promotion to a potential rival to the US-Dollar for the global reserve currency monopoly (Yinqiu & Yilin, 2019). The authoritarian Chinese regime, in contrast, appears to be enshrined for years to come. 4. Methodology
Macroeconomic dynamics such as recessions, movements in exchange rates, inflation rates or money supplies are unlikely to immediately translate into a reallocation of central bank’s foreign exchange reserves. Instead, in the eventuality of these macroeconomic dynamics central bankers globally reassess the composition of the international reserves and react accordingly. Consequently, the elaborated effects are inertial and therefore shifted forward in time. To correct for this circumstance, we use a lagged dependent variable as independent variable for the executed regression analysis as done by Chinn and Frankel in (2008) and Eichengreen et al. in (2017) before.
The sample reflects countries, whose currencies are of major importance for the investment of central banks as foreign exchange reserves. These countries are profoundly heterogenous, as few economic superpowers as the United States, China and Euroland are assessed jointly with minor economies as Switzerland, Canada and Australia nonetheless issuing strongly demanded currencies. Furthermore, the incorporation of absolute figures, as nominal GDP or absolute trade volume, in favor of relative figures (as GDP per capita or trade volume relative to GDP) cause a non-normal distribution. The original distribution of the variables was therefore highly skewed and exhibited a high kurtosis. By taking their natural logarithms the skewness was reduced to < ± 1; kurtosis has been limited to < ± 3. The distributions now rather resemble a normal distribution, allowing for an interpretation of the performed hypothesis tests. An executed Jarque-Bera-test15 underpins these findings.
As the numbers in COFER are all noted in current US-Dollars of the respective year, the figures are prone to currency price fluctuations. An appreciation of a secondary currency with respect to the US-Dollar results in an increase of its COFER share, as its price has increased. Whereas the ratio of the specific currency among the international reserves increases, the quantity of
20
reserves remains on the previous level. For the assessment of changes in the structure of the international financial system however, changes in price are of minor concern. Not changes in quantity of reserves, but a restructuring of foreign exchange reserves is of major interest. Introducing the exchange rate allows for a correction. This approach is inspired by the price change correction concept of the IMF (IMF, 2018). However, in the dataset only a change in exchange rate with respect to the US-Dollar is corrected for, as changes within two secondary currencies are only of minor concern. As these effects are only minor and correcting for price changes among two secondary currencies would have inflated needlessly, this simplified approach is used.
4.1.Relative or absolute figures
21
economies of scale come into play, usage of absolute number in favor of relative figures, seems expedient. Yet, numerous scientists previously have only taken absolute values into consideration. Finally, GDP per capita, trade volume, M2 and defense spending per GDP will be scrutinized.
4.2. Model equation and Hypothesis
The equation of the executed regression model looks as follows: Equation 1:
𝑠ℎ𝑎𝑟𝑒𝑖,𝑡 = 𝛽0+ 𝛽1𝑠ℎ𝑎𝑟𝑒𝑖,𝑡−2+ 𝛽2𝑙𝑛𝐺𝐷𝑃𝑖,𝑡 + 𝛽3𝑙𝑛𝑡𝑟𝑎𝑑𝑒𝑖,𝑡+ 𝛽4𝑙𝑛𝑀2𝑖,𝑡+
𝛽5𝑙𝑛𝑑𝑒𝑓𝑠𝑝𝑒𝑛𝑑𝑖,𝑡+ 𝛽6𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑟𝑎𝑡𝑒𝑖,𝑡+ 𝛽7𝐸𝑈𝑐𝑟𝑖𝑠𝑖𝑠𝐸𝑈,2009𝑄3−2018𝑄3 + 𝛽8 𝑓𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝑐𝑟𝑖𝑠𝑖𝑠2007−2009+ 𝛽9𝑙𝑛𝑒𝑥𝑐ℎ𝑎𝑛𝑔𝑒𝑟𝑎𝑡𝑒𝑖,𝑡+ 𝛽10𝐶ℎ𝑖𝑛𝑛𝐼𝑡𝑜𝑖,𝑡 +
𝛽11𝑙𝑛𝑖𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛𝑟𝑎𝑡𝑒𝑖,𝑡+ +𝛽12𝑖𝑛𝑡𝑒𝑟𝑒𝑠𝑡𝑟𝑎𝑡𝑒𝑖,𝑡+ 𝛽13𝑑𝑒𝑚𝑜𝑐𝑟𝑎𝑐𝑦𝑖,𝑡
Thereby, the independent variable 𝑠ℎ𝑎𝑟𝑒𝑖,𝑡 is the share of country i among the total global foreign exchange reserves in time t. The evolvement of this figures is assessed with numerous dependent variables likewise for the respective country in the respective period. As previously outlined effects will be inertial and therefore shifted in time. Therefore, 𝑠ℎ𝑎𝑟𝑒𝑖,𝑡−2 is added as
a lagged dependent variable to correct for this incident. Next, 𝑙𝑛𝐺𝐷𝑃𝑖,𝑡 displays the absolute
GDP in current US-Dollar. According to previous studies a larger economic output strengthens the position as international reserve issuer (Flandreau & Jobst, 2006; Kindleberger, 1967). l𝑛𝑡𝑟𝑎𝑑𝑒𝑖,𝑡 represents the sum of imports and exports, takes account of the significance of trade
power which has been examined previously (Eichengreen & Frankel, 1996; Papaioannou et al., 2006). 𝑙𝑛𝑀2𝑖,𝑡 incorporates the extensiveness and depth of the national financial market by
22
crisis hit the global and regional economies, confidence of specific currencies was undermined. In order to correct for these events, two dummy variables have been added to the sample. The index of Chinn and Ito (Chinnito𝑖,𝑡) is utilized to assess the openness of the capital account of a specific country, correcting for capital controls being imposed (Chinn & Ito, 2005). Expected is a rising influence as reserve issuer when the openness of the capital account increases. The inflation rate is used as a proxy to incorporate the credibility of the monetary policy of the respective central bank. Therefore, a lower inflation rate stimulates a share of the national currency among international reserves (Eichengreen, Mehl, & Chitu, 2017). The interbank interest rate is taken into consideration as a regressor. Before, strong relationships between interest rates and the status as international reserve issuer have been identified by others (Ganziro & Vambery, 2016). Lastly, the Freedom House index as a measure for the democracy in the respective nation is taken into consideration. Thereupon, open democratic economic systems stimulate the circulation of a national currency globally and consolidate the position as an international reserve issuer (Eichengreen & Frankel, 1996; Kirshner, 2008).
As other economic superpowers rise to the US, the determinants for a new potential reserve issuer are of topical interest. Concluding to a hypothesis: The COFER share of a country rises with increasing output levels, trade volume, financial depth, defense expenditures, share among vehicle currencies, openness of the capital account, sound democratic structures and diminishing inflation rates.
4.3.Model specifications:
As a first model specification, the different explaining variables are individually run in regressions. Subsequently, the sum of all regressors are executed in a pooled OLS model. Thereafter, a panel model will be used. Then the panel model will be specified once as a random and once as a fixed effects model. The subsequently performed Hausman-test16 evaluates the suitability of the two different estimators. As previously outlined the dataset contains a lagged dependent variable as additional explanatory variable. This makes the panel dynamic, which prohibits the use of fixed and random effects estimators. Therefore, the fixed and random effects test excludes the lagged dependent variable. Alternatively, a GMM estimator must be applied which allows for heteroskedasticity and the application in dynamic panel models. Afterwards and in order to increase robustness, the regressions are supported by a time fixed-effects regression. Correcting for time fixed effects however, requires dropping dummy variables for
23
common shocks, namely the global financial crisis. Further robustness checks will be executed with relative instead of absolute numbers and by dropping the smallest economies in the dataset. 5. Results & Interpretation
In the following results of the 3 different specifications will be displayed interpreted and analyzed.
5.1. Individual and Pooled OLS regressions
24
Above all, the individually executed regression analyses underline the accurate choice of different explanatory variables, as all variables reveal a strong and highly significant coefficient.
Since the exchange rate was used to absorb pricing effects of held foreign exchange reserves, an individual interpretation of this coefficient is not expedient. Alternatively, as the indicator correcting for inertia remains highly significant up to a significance level of one percent, the existence of a strong inertial effect can be safely confirmed.
Table 1: individual and pooled OLS regression
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES GDP trade Fin. Depth Vehicle Def. Spending Exchange Rate Fin. Freedom Inflation Democracy all regressors
LAGRESRATIO 0.807*** (0.0284) GDP 1.137*** 0.107** (0.0291) (0.0505) TRADEVOL 1.203*** -0.106** (0.0448) (0.0466) FIN. DEPTH 1.062*** 0.0659* (0.0522) (0.0385) VEHICLE 0.055*** 0.00234* (0.0168) (0.00141) DEFSPEND 1.056*** 0.167*** (0.0196) (0.0345) EUROCRISIS -0.0657** (0.0311) FIN. CRISIS -0.0695*** (0.0221) EXCHRAT 0.150*** -0.0244*** (0.0376) (0.00889) CHINN-ITO 0.588*** 0.118*** (0.107) (0.0286) INFLATRAT 0.407*** 0.00463 (0.0666) (0.00736) DEMOCRACY -.233*** (0.0971) -.0369228 (0.0347) INTERCEPT -15.22*** -13.32*** -6.873*** -0.939 -24.42*** 0.773*** 0.222 1.383*** 1.798*** -4.923*** (0.430) (0.555) (0.419) (0.069) (0.481) (0.0828) (0.244) (0.0764) (0.14348) (0.701) OBS 584 580 561 467 584 486 584 581 584 467 R² 0.724 0.555 0.426 0.669 0.834 0.032 0.050 0.061 0.0082 0.994
Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1; Dependent variable: ln(COFER share); Table 1 reports OLS estimates of Equation (1) based on the sample with 11 countries over the period from 1980 until 2018 Dummy variables (European debt crisis and Financial crisis) are not individually tested
All other regressed indicator’s influence remains unchanged with respect to the individually regressed indicators and in line with my predictions and previous findings. However as previously outlined, no significance can be deduced from the findings in regression (10). The results are prone to multicollinearity, yielding an excessive explanatory power. It is delusive to assume that more than 99 percent of the variance can be explained with the elaborated model. Apart from collinearity, since a lagged dependent variable has been included into the regression, the panel becomes dynamic which distorts any results yielded by a pooled OLS regression. Moreover, a diverging regression including the inertia effect (variable: 𝑙𝑛𝑙𝑎𝑔𝑟𝑒𝑠𝑟𝑎𝑡𝑖𝑜2) for every individual regression has been computed and displayed in Table 1.1.17 Although the
lagged dependent variable has been added to the individual regressions multiple indicators remain significant. Namely, all the four magnitude variables (𝐺𝐷𝑃𝑎𝑏𝑠, 𝑡𝑟𝑎𝑑𝑒𝑣𝑜𝑙, 𝑀2 and 𝐷𝑒𝑓𝑠𝑝𝑒𝑛𝑑𝑖𝑛𝑔) remain highly significant. Furthermore, the democracy indicator is highly significant but regrettably now displays a positive effect, which opposes the predicted outcome and previous findings (Kirshner, 2008).
Although the random effects estimator in Table 218 yielded higher significance than the fixed-effects estimator, a random fixed-effects model is only feasible if the variation across the different countries is random and therefore uncorrelated with the independent variable. Alternatively, if one corrects for all individual specific characteristics of all countries in the dataset over the observed time period, these characteristics can be explicitly modelled with the model. However, it is absurd to correct for all these country-specific effects over time. Instead, a fixed-effects-model should be favored in which the individual country specific characteristics are assumed to be constant over time. This finding is also reconfirmed with a Hausman-test19 yielding that
the difference in coefficients is not systematic. Therefore, the fixed-effect model is appropriate. As previously outlined the dataset is prone to collinearity complications, as the variables 𝑙𝑛𝐺𝐷𝑃𝑆𝑎𝑏𝑠, 𝑙𝑛𝑡𝑟𝑎𝑑𝑒𝑣𝑜𝑙, 𝑙𝑛𝑀2 and 𝑙𝑛𝑑𝑒𝑓𝑠𝑝𝑒𝑛𝑑 exhibit high correlations. Therefore, in Table 3 these variables are only executed individually. Likewise, the indicator for democracy and openness of the capital account exhibited a high correlation. Thus, for the computations of Table 3 these two indicators are considered individually. However, testing a dynamic panel with a fixed-effects estimator yields biased t-statistics and coefficients of determination. Consequently, a fixed effects estimator can only be employed when excluding the lagged
17 The table 1.1, displaying the output of the varying regression analysis with the inertial effect, can be found in the appendix (A.3.1).
18 Reported in the appendix (A.3.2).
27
dependent variable, which renders the dataset dynamic. As a result, we also omit the inertia in the effect. Therefore, the results must be regarded with caution.
5.2.Fixed-effects estimation (excluding inertial effect)
The computations of Table 3 include eight regressions, individually regressing the four highly correlating variables 𝑙𝑛𝐺𝐷𝑃𝑎𝑏𝑠, 𝑙𝑛𝑡𝑟𝑎𝑑𝑒𝑣𝑜𝑙, 𝑙𝑛𝑀2 and 𝑙𝑛𝑑𝑒𝑓𝑠𝑝𝑒𝑛𝑑 with the democracy and capital account openness indicator 𝑐ℎ𝑖𝑛𝑛𝑖𝑡𝑜 to avoid multicollinearity concerns. All calculations yielded a significant positive effect for the vehicle currency indicator. Thereupon an increase of the vehicle share by one percent, induces an increase of the COFER share of 0.0461 percent for the first specified model (1). Furthermore, the interest rate, Chinn-Ito and democracy indicator continuously revealed significant coefficient. Unfortunately, none of these findings are in line with the predicted outcomes and previous findings. Thereby a diminishing interest rate, stimulates a rise in the COFER share. However, as previously discussed the relevance of the interest rate for central banker’s investment decision remains questionable, as they do not pursue the intention of profit making as private investors. Instead a low interest rate, which can be synonymously seen as an indicator for discipline in the fiscal budget, could be a stimulating factor for a restructuring of foreign exchange reserves to the specific currency. The other two indicators, democracy and Chinn-Ito show a positive and negative effect respectively. This however is not in line with predictions and previous studies, as an increasing Freedom-House-index represents a less democratic structured country. Following these results, a negative influence of democratic structures could be derived for the relevance of a national currency as an international reserve currency. This seems moderately counterintuitive, as China, i.e. the only non-democratic country in the dataset, having only slight relevance as an international foreign exchange reserve issuer so far. The distortion of the coefficients most likely originates from slight deterioration of the democracy index of Western countries, which in comparison to China still dispose flawless democracies. Once again suboptimal are the stagnant figures of the democracy indicator, which rarely change across time.
However, since a fixed-effects estimator has been employed and both variables are almost stagnant over time, it is likely that these effects are captured by the country-fixed effects, which is verified at a later stage.
28
The last two regressions included lndefspend and the democracy and capital account openness indicator as explanatory variable individually and all other variables, displayed in column (7) and (8). In the former the variable for defense spending revealed a significant coefficient at a significance level of 5 percent. Hence, a positive relationship was detected, according to which an increase in defense budget of one percent stimulates the COFER share by 0.186 (7) or 0.24 (8) percentage points. It consequently affects the significance of the national currency in the role as an international reserve currency positively. This finding complies with the predicted outcome and previous findings (Kirshner, 2008).
All other variables show similar outcomes as in earlier specifications. Although, once more the coefficients of lnexchrat are highly significant, any deduction is not advisable, as this variable only absorbs price changes in the COFER figures.
Table 3: Fixed effects estimators without lagged dependent variable as explanatory variable
(1) (2) (3) (4) (5) (6) (7) (8)
VARIABLES GDP GDP Trade Trade M2 M2 Defspend Defspend
lnGDPabs -0.0110 -0.0291 (0.271) (0.315) lntradevol -0.0721 -0.0570 (0.115) (0.132) lnM2 0.221 0.238* (0.119) (0.122) lndefspend 0.186* 0.240** (0.0791) (0.0895) vehiclerate 0.0461*** 0.0325** 0.0458*** 0.0328*** 0.0433** 0.0290** 0.0422** 0.0285** (0.0118) (0.00949) (0.0101) (0.00757) (0.0143) (0.00929) (0.0134) (0.00993) EUcrisis -0.0686 -0.0769 -0.0432 -0.0579 -0.129 -0.146 -0.0941 -0.113 (0.144) (0.145) (0.154) (0.154) (0.105) (0.100) (0.111) (0.103) lnexchrat 0.679** 0.630* 0.675** 0.634** 0.542** 0.486** 0.674** 0.633** (0.231) (0.261) (0.232) (0.256) (0.172) (0.166) (0.228) (0.247) Financrisis -0.111 -0.0895 -0.0974 -0.0808 -0.161 -0.144 -0.142 -0.132 (0.101) (0.103) (0.107) (0.108) (0.104) (0.106) (0.111) (0.110) lninflatrat 0.00486 0.0124 0.00499 0.0119 -0.00485 0.00236 0.00290 0.00984 (0.0207) (0.0186) (0.0216) (0.0186) (0.0138) (0.0135) (0.0232) (0.0217) lninterestrate -0.0929*** -0.0882*** -0.0974** -0.0894** -0.0606*** -0.0522*** -0.0786** -0.0702* (0.0221) (0.0181) (0.0364) (0.0343) (0.00760) (0.00664) (0.0314) (0.0311) democracy 0.279*** 0.279*** 0.281*** 0.253*** (0.0580) (0.0349) (0.0373) (0.0245) chinnito -1.691** -1.621** -1.850*** -1.718*** (0.541) (0.442) (0.259) (0.304) Constant -0.883 3.911 -0.152 4.010*** -2.583** 2.201*** -5.454** -2.200 (4.020) (3.440) (1.461) (0.809) (0.743) (0.498) (1.843) (1.678) Observations 553 553 549 549 553 553 553 553 R-squared 0.340 0.341 0.350 0.348 0.390 0.398 0.354 0.364 Number of countries 11 11 11 11 11 11 11 11
Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1; Dependent variable: ln (COFER share);
5.3.Dynamic model estimation (including inertial effect)
Table 4: Arellano -Bover & Blundell-Bond estimator
(1) (2) (3) (4) (5) (6) (7) (8)
VARIABLES GDP GDP Trade Trade M2 M2 Defspend Defspend
Lnlagresratio2 0.761*** 0.767*** 0.900*** 0.899*** 0.866*** 0.872*** 0.687*** 0.689*** (0.0361) (0.0355) (0.0269) (0.0269) (0.0305) (0.0297) (0.0426) (0.0421) lnGDPabs 0.320*** 0.308*** (0.0499) (0.0484) lntradevol 0.130*** 0.135*** (0.0405) (0.0406) lnM2 0.212*** 0.196*** (0.0533) (0.0500) lndefspend 0.325*** 0.321*** (0.0464) (0.0456) vehicle rate 0.000589 0.00101 0.00432 0.00413 0.00594* 0.00613* 0.00636** 0.00652** (0.00343) (0.00340) (0.00362) (0.00360) (0.00354) (0.00354) (0.00325) (0.00324) Eurocrisis 0.143** 0.143** 0.135** 0.133** 0.0257 0.0361 0.149*** 0.148*** (0.0590) (0.0592) (0.0646) (0.0646) (0.0722) (0.0713) (0.0573) (0.0573) Financial crisis -0.0779** -0.0704** -0.0663* -0.0651* -0.0877** -0.0807** -0.106*** -0.0984*** (0.0331) (0.0330) (0.0358) (0.0356) (0.0360) (0.0356) (0.0330) (0.0326) lnexchrat -0.00503 0.0138 0.0190 0.0258* -0.0298* -0.0155 0.0203* 0.0403*** (0.0112) (0.0119) (0.0117) (0.0133) (0.0160) (0.0145) (0.0105) (0.0122) lninflatrat 0.00485 0.00566 0.00875 0.00856 0.00965 0.0101 0.0104 0.0110 (0.0122) (0.0122) (0.0134) (0.0134) (0.0129) (0.0129) (0.0117) (0.0118) lninterestrate 0.0394*** 0.0391*** 0.0201 0.0212 0.0179 0.0174 0.0225* 0.0234** (0.0129) (0.0130) (0.0135) (0.0136) (0.0129) (0.0129) (0.0117) (0.0118) Democracy -0.107*** -0.0369 -0.0652** -0.117*** (0.0260) (0.0242) (0.0275) (0.0255) chinnito 0.184*** 0.0528 0.121** 0.194*** (0.0439) (0.0399) (0.0481) (0.0425) Constant -4.866*** -4.176*** -1.724*** -1.621*** -1.899*** -1.433*** -8.148*** -7.492*** (0.760) (0.644) (0.537) (0.464) (0.478) (0.348) (1.164) (1.051) Observations 546 546 542 542 546 546 546 546 Number of country 11 11 11 11 11 11 11 11
Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1; Dependent variable: ln(COFER share);
Table 4 reports Arellano-Bover & Blundell-Bond estimates of Equation (1) based on the sample with 11 countries over the period from 1980 until 2018 including the lagged dependent variable correcting for inertia. Number of instruments(lags): 171 Sargan-Test (Appendix: A.4.7) revealed an appropriate number of instruments.
To be more precise, taking China and the US as examples: if the Chinese capital account is as unrestricted as the US-American one, its COFER share rises by approximately 0.5 percentage points. If China was as democratic as the other Western countries in the dataset, according to the calculations in Table 4, its COFER share would rise by 0.7 percentage points.
The third (3) and fourth (4) regression of table 4 included the trade volume of the specific reserve currency issuer. This variable proves to be highly significant in explaining the evolvement of COFER figures over time. Specifically, an increase in the trade volume figures of one percent, stimulates the COFER figures positively by 0.13 percentage. Beyond that, few other coefficients are significant, to a lower extent than in regressions (1) and (2).
Regressions (5) and (6) of table 4 included the financial depth, assessed by means of the money supply (M2) of the respective country. For this explanatory variable, the regression revealed highly significant estimators. Whereupon an increase in the money supply of 1 percent, stimulates the COFER shares by about 0.2 percent. Moreover, for these two regressions the estimator for vehicle rate exhibits a high significance with a small positive effect. Thus, if the trade with a national currency is intensified in foreign exchange markets, its relevance will rise accordingly. This confirms previous findings, according to which the relevance of a national currency as foreign exchange reserve employed by other nations, rises with increased use in international settlements and foreign exchange trading (Gopinath & Stein, 2018). Again, the coefficients for the financial crisis, democracy and openness of the capital account exhibit a significant effect. Regrettably, the coefficient for the European debt crisis again counterintuitively indicates a positive effect.
33 6. Robustness
The data set revealed several challenges to perform econometric analyses with robust and valid results. Firstly, as numerous figures in the dataset were prone to heteroskedasticity, the use of robust standard errors was inevitable. Yet, the employment of robust standard errors blows up standard errors, which reduces the significance of estimators and therefore decreases the overall coefficients of determination. Moreover, multiple variables which are individually all significant in explaining the relevance of a national currency as international reserve currency disposed high correlation among themselves. Regressing them jointly, arises a multicollinearity problem. Therefore, evidential value cannot be deducted from these regressions. For the confirmation of the hypotheses of this work, table 3 and 4 have to be consulted, as they offer the soundest results. Whereas the coefficients of determination are defectively overblown in the first two tables, the results of the latter two, in particular table 4 are viable, as it corrects for inertia.
6.1.Robustness check with time-fixed effects
To reconfirm the robustness of those two tables, we rerun their regressions including time-fixed-effects to further increase their validity. First, the fixed regression estimations of table 3 will be repeated with time-fixed effects. Whereas the previous estimations only corrected for country specific effects, the time-fixed effects estimator also corrects for unobservable characteristics which change over time. This estimation adds a dummy variable for every single time period, which captures year-specific effects. Subsequently, the dynamic estimations of table 4 will be replicated with time-fixed-effects as well.
Table 3.120 added time-fixed effects to the performed regressions of Table 3. Above all, the results of the fixed and time-effects estimators can confirm the deductions made from the results displayed in table 3. Additionally, other explanatory variables now reveal a significant effect. The new regressions exhibit a significant positive effect for GDP levels, financial depth and defense spending. Likewise, the dummy variable for the European debt crisis shows significance coefficients and is being reconfirmed. Unlike in Table 3, in which an increasing inflation rate supported the internationalization of the national currency, the calculations in table 3.1 revealed a negative coefficient, according to which a lower interest rates stimulates the COFER share positively. As the time-fixed effects model yields significant values for inflation rate, this specification therefore indicates that the effect of inflation rates changed
34
considerably over time. This deduction is in line with nominal data, as inflation rates stabilized within the last 20 years. As outlined previously, inflation rates have been almost stagnant, therefore these results are highly questionable. These doubts now have been confirmed by totally diverging results in two different analyses. Regrettably, the indicator for the openness of the capital account chinnito this time also yields a negative coefficient. This is rather counterintuitive, as it opposes previous findings and the hypothesized prediction. The time-fixed effects estimator takes changes of the openness of capital accounts (for instance: liberalizations of capital controls) across time into consideration. When observing the actual numbers, COFER rate of, for instance, China have been rising, although they have been performing weakly in the Chinn-Ito index.
Additionally, since the country fixed-effects estimator might capture the effects of the relatively stable variables democracy and capital account openness, this regression has been repeated excluding these two variables. The results are shown in table 3.1.1.21 However, the test results in a loss of significance and coefficient of determination. Therefore, the fixed-effects is unable to cover the two omitted variables.
The time-fixed effects have been tested for joint significance, which proved to be significant at a significance level of one percent. Therefore, time fixed effects are present, and its employment further improves our model. The fixed effects estimation which also corrects for changes across time exhibits a higher R². Above all, it reconfirms the major conclusions which have been drawn from initial Table 3.
The robustness of Table 4 as the soundest analysis, correcting for inertia, heteroskedasticity and serial correlation is verified by adding time fixed effects. Since time-fixed effects are applied, dummy variables for common shocks must be removed. Table 4.122 displays the estimators for
the modified regression. As only minor changes materialize, the new estimations including time-fixed effects reconfirm the findings of the initial calculations. The most noticeable change is the loss in significance for the trade indicator when time-fixed effects are included into the model. Most likely, a large share of explanatory power of trade volume can also be explained by year specific developments, which are different over time but not for countries. For instance, year specific declines in trade figures caused by global recessions cause a reduction in explanatory power, as it is now captured by the time-fixed effects. Moreover, the effects for the openness of the capital account and democratic structures remain unchanged. Similarly, the
35
European debt crisis dummy once more revealed a delusive positive effect, which opposes the initial data. Although, only common shocks must be removed in a time-fixed effects estimation, as it considers these common shocks. The European debt crisis however only affected the Eurozone and could therefore hardly be covered by the time fixed effects. Generally, the original estimations from Table 4 proved to be viable, as correcting for time-fixed effects only slightly changes the results.
6.2.Robustness check with relative indicators
Although previous work has emphasized the relevance of magnitude effects for the determination of a reserve currency issuer (Cohen, 2015; Norrlof, 2014), as an additional robustness check the four magnitude variables (GDPabs, M2, tradevol and defspend) which all revealed positive effects on COFER share with high significance levels, these estimations are reran with relative figures of these indicators. Whereas for GDP, the GDP per capita is considered, the other three variables are put into relation as a share of the absolute GDP. With these variables the executions of Table 3 and 4 as the soundest estimation will be reran. Table 3.223 displays the results from the original Table 3 which has been re-estimated with relative variables. Apart from the variable M2/GDP no other magnitude variable reveals an estimator significantly different from zero. Thereby, an increase of M2 proportionated to GDP of one percent induces a rise in COFER share of 0.5 percentage points. Whereas in the initial Table 3 the defense spending exhibited a significant positive effect, this effect ceases when using a relative measure. However, using the M2 indicator as a proportion of the absolute GDP level is problematic, unless in a crisis when output levels sink, and money supply rises due to crises intervention practices, as both figures rise normally, rendering a new stable variable. All the other estimators stayed the same.
Equally, the estimations of Table 4 with the Arellano -Bover & Blundell-Bond estimators were retaken with relative magnitude variables. This estimation includes an inertia-correcting lagged dependent variable. The corresponding results can be found in the appendix in Table 4.2.24 The
results are disillusioning, as they reveal results which are totally inconsistent with previous findings and predictions. Whereas the previous model specification revealed a positive effect for M2 per GDP, the effect shifted negative now. Similarly, the calculation for trade volume relative to GDP also revealed a negative effect. Although this outcome is opposing the predictions, there is a good explanation for it. As previously outlined, due to hub effects or
